Discrete and Continuous Time Markov Properties

86 Views Asked by At

$\newcommand{\indep}{\perp\!\!\!\perp}$ In discrete time, the Markov property is $$P[X_{n+1}\in A\mid X_n=s_n,X_{n-1}=s_{n-1}\dots ]=P[X_{n+1}\in A\mid X_n=s_n]$$ On the other hand, the "general Markov property" as given in Kallenberg's Foundations of Moderns Probability is the conditional independence $$X_u \indep _{X_t}{} \mathcal{F}_t$$ for any $t\leq u$.

What's bugging me is the freedom of $u$ and $t$ in the general case compared to $n+1$ and $n$ in the discrete time - in the general case one could take $u$ much larger than $t$, yet there is nothing like $$P[X_{m}\in A\mid X_n=s_n,X_{n-1}=s_{n-1}\dots ]=P[X_{m}\in A\mid X_n=s_n]$$ for $m\geq n$.

Why is the discrete time formulation as it is?

1

There are 1 best solutions below

4
On

The version of the Markov property in discrete time that you stated in your post implies the more general property that, for every $k\geqslant1$, $$P[X_{n+k}\in A\mid X_n=s_n,X_{n-1}=s_{n-1},\ldots,X_0=s_0]=P[X_{n+k}\in A\mid X_n=s_n].$$ Equivalently, for every $m\geqslant n+1$, $$X_m \indep_{X_n}\mathcal{F}_n.$$